Experimental Evolution of Aging, Growth, and Reproduction in Fruitflies

Experimental evolution of aging, growth, and reproduction in fruitflies

S. C. Stearns*, M. Ackermann, M. Doebeli, and M. Kaiser

Zoology Institute, University of Basel, Rheinsprung 9, CH-4051 Basel, Switzerland

Edited by May B. Berenbaum, University of Illinois at Urbana-Champaign, Urbana, IL, and approved January 24, 2000 (received for review July 12, 1999) We report in this paper an evolutionary experiment on Drosophila flies (19–22). It succeeds because evolution is more rapid than that tested life-history theory and the evolutionary theory of until recently was expected (23, 24). aging. As theory predicts, higher extrinsic mortality rates did lead Previous artificial selection experiments on Drosophila (25– to the evolution of higher intrinsic mortality rates, to shorter 31) suggested that the evolution of lifespan is constrained by lifespans, and to decreased age and size at eclosion; peak fecundity genetic links between early-life traits, such as early fecundity and also shifted earlier in life. These results conﬁrm the key role of larval competitive ability, and late-life mortality and fecundity extrinsic mortality rates in the evolution of growth, maturation, rates. The longest running artificial selection experiment on the reproduction, and aging, and they do so with a selection regime evolution of aging in fruit flies (25) started in 1976. One set of that maintained selection on fertility throughout life while holding lines is only allowed to reproduce early in life, another only late population densities constant. in life; fecundity and survival are the targets of selection, and the evolutionary responses in traits such as intrinsic mortality are life-history evolution ͉ lifespan ͉ age at maturity ͉ body size ͉ Drosophila measured. The late-reproducing lines have evolved a much longer lifespan and lower intrinsic mortality rates than the early-reproducing lines. This experiment has been repeated in n this paper, we report a case study in experimental evolution another laboratory with a design that avoids unintentional Iwith the fruit fly Drosophila melanogaster that is designed to selection and controls larval density (31). In this second case, the test predictions of life-history theory (1–6) and the evolutionary late-reproducing lines have improved survival and decreased theory of aging (7–11). It did confirm those predictions. The early-life fertility, but there has been no improvement in late-life change in the environment that drove phenotypic change was a fertility, and no correlated responses to selection in the preadult difference in extrinsic adult mortality rates. What is different in period have been observed. In the best field test of life-history this experiment is that mortality was applied in a way closely theory, still continuing in Trinidad (18, 24), extrinsic mortality resembling natural conditions rather than by using traditional rates of guppies are manipulated by exposing them to different artificial selection. Treatments differed only in adult mortality predators. Significant genetically based changes in age at matu- applied twice each week, which maintained selection on fertility rity and fecundity in the predicted direction have been observed, throughout life. and changes in lifespan are expected.

Evolutionary Theory of Aging and Life Histories. Evolutionary theory Our Experiment: High Versus Low Adult Mortality. Our experiment predicts the impact of a difference in extrinsic mortality on subjected a stock of wild D. melanogaster to two treatments: high adult mortality (HAM) and low adult mortality (LAM). In intrinsic mortality rates (hence, lifespan) and on growth, matu- contrast to previous experiments, we knew age-specific extrinsic ration, body size, and reproduction. When extrinsic mortality mortality precisely, because we imposed it twice weekly. Thus, rates increase, they lower the probability of survival to a given the problem posed by selection to the organisms differed from age and cause the strength of selection to decline faster with age, the problems posed by previous work (25, 31), because the LAM making an increase in intrinsic mortality rates with age ‘‘more flies were contributing offspring to future generations through- affordable’’ or ‘‘less avoidable.’’ From this concept follows a out their lives, not just at the ends of their lives. This selection central prediction of the evolutionary theory of aging (11): regime, which maintains selection on fertility throughout adult Higher extrinsic mortality rates should lead to higher intrinsic life, is a much closer approximation of the kind of selection that mortality rates and a decrease in lifespan, which is a prediction is likely to be found in nature than is selecting for flies that lay adumbrated by Weismann (7) and Medawar (8), explicit in the eggs only very early or very late in life (25, 31). Moreover, work of Williams (9), quantitative in the research of Hamilton because the population density of both the adults (200 per (10) and Charlesworth (1), and consistent with comparative population cage) and the larvae (250 per unit of larval medium) evidence (11–13). was the same in both treatments, the responses can be attributed Extrinsic mortality rates also affect the evolution of other solely to differences in mortality rates with no confounding life-history traits. Higher extrinsic adult mortality rates should effects of larval or adult density. lead to higher reproductive effort early in life and, for age- but We checked that the predictions of both the evolutionary not stage-dependent life histories (5), to more rapid develop- theory of aging and the reproductive-effort model held for our ment and eclosion at an earlier age and a smaller size (1–6). experimental conditions by analyzing computer models simulat- ing the evolutionary dynamics of phenotypes endowed with Experimental Evolution and Artificial Selection. Recently, a new tool has been exploited to test such predictions: experimental evo- This paper was submitted directly (Track II) to the PNAS ofﬁce. lution (14). In contrast to artificial selection, in which the Abbreviations: HAM, high adult mortality; LAM, low adult mortality. experimenter determines which trait is selected, in experimental *To whom reprint requests should be addressed. E-mail: [email protected]. evolution, the experimenter creates the conditions under which The publication costs of this article were defrayed in part by page charge payment. This a prediction should hold and lets the evolving population article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. determine with which traits the problem will be solved. This §1734 solely to indicate this fact. EVOLUTION approach has yielded important insights with bacteria (14–16), Article published online before print: Proc. Natl. Acad. Sci. USA, 10.1073͞pnas.060289597. algae (17), poeciliid fish (18), and temperature adaptation in Article and publication date are at www.pnas.org͞cgi͞doi͞10.1073͞pnas.060289597

PNAS ͉ March 28, 2000 ͉ vol. 97 ͉ no. 7 ͉ 3309–3313 different life-history traits and age schedules of intrinsic mortalities. In the simulations, these phenotypes were subjected to selection regimes with different adult mortalities that corre- sponded to the regimes used in the experiment. The models assumed the standard constraining relations among development time, body size, and fecundity that are known to Drosophila researchers (5). The prediction that the HAM treatment should evolve higher intrinsic mortalities held for a range of assumptions about the shapes of intrinsic-mortality curves, including Gompertz (32) and sigmoid. The predictions of the reproductive- effort model also were confirmed. The computer models predicted that HAM flies would evolve shorter development times, smaller size at eclosion, higher fecundity early in life, and lower fecundity later in life than did LAM flies. These expectations have the following intuitive explanation. In this experiment, extrinsic adult mortality starts when flies enter a population cage at the age of 14 days from egg. Under the high mortality in the HAM treatment, most flies can only reproduce for a few days before they are killed; they should evolve high fecundity at the age of 14 days by eclosing early and by being at the peak of their age-fecundity curve when they enter the cage. In the LAM treatment, most flies can reproduce for several weeks; fecundity late in life is important and can be increased by eclosing later at a larger size, which, however, delays maturity and lowers fecundity at age 14 days. Note that if the shift in mortality rates occurs precisely at eclosion, rather than at a particular age, then an increase in adult mortality selects for a longer juvenile period (at low risk) and a larger size with higher fecundity at eclosion. Methods The founder stock of 820 flies consisted of 10 virgin males and 10 virgin females from each of 41 isofemale lines that were collected in and near Basel, held as isofemale lines for several years, then bred together in a single cage for about five gener- Fig. 1. Intrinsic and total mortality in the population cages. Intrinsic mor- ations to yield a genetically variable population. To found the tality is indicated by the irregular lower lines, one line per replicate. Negative mortalities indicate that too many flies accidentally were added in the last selection lines, we collected eggs once a week for 5 weeks to get check. Total mortality (intrinsic plus extrinsic mortality) is given by the thick five age classes and to avoid the cycles of reproduction and age straight line. The symbols indicate when there were not enough flies available classes that persist in populations founded with flies of one age. to achieve the target mortality because not enough eggs had been laid 14 days The experiment started November 15, 1993, is continuing, and previously. ᮀ, Replicate 1; E, replicate 2; ‚, replicate 3. Filled symbols indicate is maintained on a day–night cycle of 12 h of light, 25°C͞12hof females, and open symbols indicate males. (Upper) HAM. The symbols above dark, 20°C, and at 70% relative humidity. Eggs for recruits are the total-mortality line indicate that 4 times a replicate was lost because of collected from the population cages on Tuesdays and Fridays, at mislabeling and was replaced from backup stocks 2 weeks behind the current which time two bottles for each line are established, each with history of selection. At the start of the experiment and after larval density was 250 eggs in a fixed amount of larval medium. The adults that increased in January 1996, there were sometimes not enough replacement flies to achieve the target mortality (23 times for females, 2.1% of the checks; hatch from these eggs are used 14 days later to replace flies that 20 times for males, 1.8% of the checks). (Lower) LAM. Before January 1996, die or are killed in the population cages. natural mortality rates twice exceeded 20%. After January 1996, when the Each treatment consists of three replicates, and each replicate target mortality was 10%, there were more old flies in the cages, and intrinsic has 100 male and 100 female adults, enough to avoid inbreeding. mortality increased. From January 1996 to July 1998, intrinsic mortality was Adult mortality rates are imposed and densities are reestablished 11.3% for females (vs. 2.8% in HAM) and 10.3% for males (vs. 3.6% in HAM). by hand with 14-day-old flies on Mondays and Thursdays (Fig. 1); larvae are maintained at the same density in the same medium for both treatments. We could maintain constant larval week as an adult to P ϭ 0.81. Those changes increased both the densities because of excess production of eggs and constant adult physiological stress on the larvae and the difference between the densities, because we reared more flies than were needed. two mortality regimes. We have achieved the target adult Extrinsic mortality was adjusted so that intrinsic plus extrinsic mortality rates in the two treatments (Fig. 1). mortality reached the target level. Traits were assayed in vials, not in the population cages The mortality rates imposed and the rearing conditions have themselves, in measurements made parallel to the main selection changed twice since the start of the experiment. In the first 13.5 experiment. For each of the three replicates in both selection months in the HAM treatment, 90% of the flies in the cage were regimes, 4 ϫϷ50 14-day-old flies (age measured from egg) were killed and replaced twice per week; the probability of surviving collected in 250-ml glasses and fed fresh yeast for 4 h, then 1 week as an adult was P ϭ 0.01. In the LAM treatment, the allowed to lay eggs into Petri dishes for 3 h. About 26 h after egg probability of surviving 1 week as an adult was P ϭ 0.64. In laying, 10 vials with 25 larvae each were set up for each replicate. January 1995, larval density was raised from 6.25 to 10.40 larvae One day before eclosion began, the vials were positioned in per ml, and food quality was lowered from 1% to 0.75% yeast in random order in a machine that collected newly hatched flies at the larval medium. In January 1996, larval density was raised to 2-h intervals. After eclosion (8–12 h) males and females were 12.50 larvae per ml, and the adult mortality rate in the LAM separated. Males were frozen, dried for3hat50°C, and weighed treatment was lowered, raising the probability of surviving 1 to Ϯ 1 ␮g. Females were weighed alive (Ϯ10 ␮g) and set up in

3310 ͉ www.pnas.org Stearns et al. Fig. 2. Changes in dry weight (a), time to eclosion (b), and early fecundity (c), 1993–1997. *, P Ͻ 0.05; **, P Ͻ 0.01; ***, P Ͻ 0.001. The bars at the end of the series in a and b are the standard errors of three sets of measurements made in the spring of 1998. Divergence between HAM and LAM (LAM minus HAM) in dry weight (d), time to eclosion (e), and early fecundity (f), 1993–1997.

ϫ individual vials (52 18 mm) with two males apiece of the same changed every day. Eggs were counted automatically with im- EVOLUTION age and line to determine fecundity. The medium in the vial caps aging software until 95% of the females had died; dead and was covered with 1:10 (vol͞vol) diluted fresh yeast; caps were trapped males were replaced until all females had died.

Stearns et al. PNAS ͉ March 28, 2000 ͉ vol. 97 ͉ no. 7 ͉ 3311 Table 1. Variation among replicates and differences between treatments HAM LAM P values

Treat͞ Repl.͞ Trait n Repl. 1 Repl. 2 Repl. 3 Total n Repl. 1 Repl. 2 Repl. 3 Total repl., ms* error, ms*

Females Development time, h 389 252 255 256 254 345 268 271 277 272 0.0041 0.0009 Dry weight, ␮g 90 250 236 240 242 90 257 261 265 261 0.0156 0.0468

Fecundity Days 13–15 340 43.8 39.7 38.8 40.8 322 28.5 28.1 24.8 27.0 0.0035 0.1071 Days 31–33 (late) 124 31.7 29.9 26.2 29.8 104 31.9 31.3 35.1 33.2 0.1008 0.7189

Males Development time, h 389 259 259 261 260 334 270 280 277 276 0.0061 0.0001 Dry weight, ␮g 388 206 192 194 197 332 221 212 217 217 0.0182 0.0054

Mean values of traits measured in November 1997, 4 years after the experiment started, except female dry weights, which were measured February 1998. Repl., replicate; ms, mean square. *Signiﬁcance levels from ANOVA.

Intrinsic mortalities were measured precisely in 1998, when as predicted (Table 1). The HAM flies developed more rapidly, the HAM treatment had experienced Ϸ90 and the LAM treat- eclosed at a smaller size, and reached peak fecundity more ment Ϸ50 generations, on 5,000 flies of each sex for each rapidly. Then the dynamics slowed, either because the lines had replicate in two cages per replicate with 2,500 males and 2,500 reached new adaptive peaks or because genetic variation for females per cage (5,000 ϫ 2 sexes ϫ 3 replicates ϫ 2 treat- improvements was exhausted by selection and inbreeding. ments ϭ 60,000 flies). Dead flies were sexed, counted, and The results qualitatively confirmed the assumptions we made replaced every 3 days with white-eyed mutants to hold the in the computer model: Flies that eclosed earlier were lighter density constant throughout the assay. At the end, the number (r ϭ 0.86; P ϭ 0.02) than flies that eclosed later and had higher of flies remaining in each cage was counted to estimate the early fecundity (day 13–15 after egg; r ϭ 0.99; P Ͻ 0.0001, number of escapees, which averaged about 4% of the flies. We calculated with the six-line means). distributed escapes uniformly over age classes in proportion to Intrinsic mortality was significantly higher for HAM than for the number of flies surviving. LAM flies from the 40th to the 75th day of life (Fig. 3). Thus, Fecundity, development time, and weight were analyzed by differences in extrinsic mortality rates did lead to the evolution ANOVA [SAS Institute (Cary, NC) procedure GLM] with of the expected differences in intrinsic mortality rates. Recently, replicate lines nested within treatments and, where appropriate, there has been considerable discussion (33–36) about the shape replicate vials nested within lines. Line MS was used to test of mortality curves, stimulated by the observation of decelerat- treatment effects. Mortality curves were analyzed with Cox ing mortalities late in life (37). In this study, female age-specific regressions (SAS procedure PHREG) with the same design as intrinsic mortality (d ) stopped increasing after about 65 days of the ANOVA. Data were right-censored for escapees. x age, then decreased, then increased again (Fig. 3), suggesting Results and Discussion that mortality rates do slow their rate of increase and may The female flies responded in the first year of their new life in decrease late in life. population cages by taking longer to develop, becoming smaller at eclosion, and decreasing slightly in early fecundity (male response was similar), regardless of the treatment. Time to eclosion increased in both treatments from 266 to 285 h. Dry weights at eclosion held constant at about 380 ␮g. Early (days 13–15) fecundity dropped from Ϸ75 eggs per day to Ϸ50 eggs per day (Fig. 2 a–c). The assay methods were improved during the first year of the experiment, and the changes in the absolute values of life-history traits during the first year can mostly be attributed to changes in the assay methods. Divergence between the two treatments in time to eclosion and dry weight began after 1 year when larval food quality was lowered and larval density was increased; divergence in early fecundity began after 2 years when larval density was raised again and the adult mortality rate in the LAM treatment was lowered. In all three traits, divergence continued for 1–2 years, then slowed considerably or stopped (Fig. 2 d–f). The slow change at the beginning of the experiment can be attributed to two factors: the larvae were not food-stressed enough to express the genetic variation that allows a response to selection, and the difference in the adult mortality rates was not yet large enough to provoke a rapid response. After the stress on the larvae was Fig. 3. Intrinsic mortality rates per 3-day interval. Number dying in each increased in 1995 and again in 1996, and the difference in period determines statistical power. Results show averages for the two mortality rates was increased in 1996, the lines diverged rapidly, treatments. Mortality curves were measured April–July 1998.

3312 ͉ www.pnas.org Stearns et al. We rarely have measured a significant difference between treatments in late fecundity in 5 years of yearly assays. In the HAM treatment, late fecundity is not under selection, but may be subject to reduction through mutation accumulation. In the LAM treatment, late fecundity is under direct selection in all survivors. The fact that late fecundity has not decreased more dramatically in the HAM flies suggests that the effects of mutation accumulation, if any, or of an antagonistic pleiotropic connection with early fecundity, if any, are not strong. Thus, our results do not support the existence of a genetic connection between early and late fecundity as suggested by Rose (25) but not by Partridge et al. (31). Was there a connection between fecundity and subsequent mortality, a cost of reproduction? In 1996, we checked for a relationship between early fecundity and subsequent adult mortality by calculating the correlation of early fecundity with mortality rates defined on subsequent 5-day intervals (Fig. 4). Fig. 4. Correlations of early fecundity with subsequent age-speciﬁc mortal- With measurements on only six lines, power was low, but ity deﬁned on 5-day age classes. Open symbols, HAM; closed symbols, LAM. correlations were consistently strongly positive for about 2 weeks Traits were measured in November 1996. after the fecundity measurement and, for one age class, they were significantly so. Thus, we confirmed the positive relation- This experiment tested and confirmed the prediction that ship between early fecundity and subsequent mortality found by higher extrinsic mortality rates lead to the evolution of higher Rose (25) and Partridge et al. (31) that is a cornerstone of the intrinsic mortality rates, something that had been done previ- life-history theory and the evolutionary theory of aging (5, 6, 11). ously (25–31), but never before with such a realistic and precisely However, in our case, the strength of the correlations arises measured difference in mortality regimes and with such precise because the HAM lines had (i) high early fecundity and (ii) high control of both adult and juvenile population densities. It also confirmed all of the major predictions of life-history theory in a later mortality, whereas the LAM lines had (iii) low early case in which mortality rates were known precisely and where the fecundity and (iv) low later mortality, a pattern explicable as confounding effects of density variation in both adults and larvae direct responses to selection by both traits in three (i, iii, iv)of were excluded. The interaction of extrinsic mortality rates with four cases. Only the relatively high late mortality of the HAM intrinsic constraining relations was sufficient to explain these flies (ii) is not selected directly and suggests either a link to early results. fecundity, decreased maintenance, or effects of mutation accumulation. Our results do not contradict stronger evidence from We thank Mark Drapeau, Dieter Ebert, Tad Kawecki, Nancy Knowlton, Tom Little, Linda Partridge, and three anonymous referees for improve- other experiments (e.g., ref. 38) in which the relationship ments to the manuscript, and Anni Mislin, Hanni Zingerli, and Barbara measured could not have been caused by direct responses to Sykes for technical support. This work was supported by the Swiss selection. National Science Foundation.

1. Charlesworth, B. (1980) Evolution in Age-Structured Populations (Cambridge 21. Huey, R. B., Partridge, L. & Fowler, K. (1991) Evolution 45, 751–756. Univ. Press, Cambridge, U.K.). 22. Partridge, L., Barrie, B., Barton, N. H., Fowler, K. & French, V. (1995) 2. Gadgil, M. & Bossert, W. (1970) Am. Nat. 104, 1–24. Evolution 49, 538–544. 3. Schaffer, W. M. (1974) Ecology 55, 291–303. 23. Gingerich, P. D. (1983) Science 222, 159–161. 4. Le´on, J. A. (1976) J. Theor. Biol. 60, 301–335. 24. Reznick, D. N., Shaw, F. H. & Shaw, R. G. (1997) Science 275, 1934–1936. 5. Roff, D. A. (1992) The Evolution of Life Histories (Chapman & Hall, London). 25. Rose, M. (1984) Evolution 38, 1004–1010. 6. Stearns, S. C. (1992) The Evolution of Life Histories (Oxford University Press, 26. Luckinbill, L., Arking, R., Clare, M.-G., Cirocco, W. C. & Buck, S. A. (1984) Oxford). Evolution 38, 996–1003. 7. Weismann, A. (1889) Essays upon Heredity and Kindred Biological Problems 27. Partridge, L. & Fowler, K. (1992) Evolution 46, 76–91. (Clarendon, Oxford). 28. Hillesheim, E. & Stearns, S. C. (1992) Evolution 46, 745–752. 8. Medawar, P. B. (1952) An Unsolved Problem in Biology (Lewis, London). 29. Zwaan, B. J., Bijlsma, R. & Hoekstra, R. F. (1995) Evolution 49, 635–648. 9. Williams, G. C. (1957) Evolution 11, 398–411. 30. Leroi, A. M., Chippendale, A. K. & Rose, M. R. (1994) Evolution 48, 10. Hamilton, W. D. (1966) J. Theor. Biol. 12, 12–45. 1244–1257. 11. Rose, M. (1991) Evolutionary Biology of Aging (Oxford Univ. Press, Oxford). 31. Partridge, L., Prowse, N. & Pignatelli, P. (1999) Proc. R. Soc. Lond. B 266, 12. Keller, L. & Genoud, M. (1997) Nature (London) 389, 958–960. 13. Ricklefs, R. E. (1998) Am. Nat. 152, 24–44. 255–261. 14. Bennett, A. F. & Lenski, R. W. (1999) Am. Zool. 39, 346–362. 32. Gompertz, B. (1825) Philos. Trans. R. Soc. London A 115, 513–525. 15. Elena, S. F., Cooper, V. S. & Lenski, R. E. (1996) Science 272, 1802–1804. 33. Abrams, P. A. & Ludwig, D. (1995) Evolution 49, 1055–1066. 16. Bennett, A. F. & Lenski, R. E. (1997) Evolution 51, 36–44. 34. Mueller, L. D. & Rose, M. R. (1996) Proc. Natl. Acad. Sci. USA 93, 17. Bell, G. (1997) Heredity 78, 490–497. 15249–15253. 18. Reznick, D. N., Bryga, H. & Endler, J. A. (1990) Nature (London) 346, 357–359. 35. Charlesworth, B. & Partridge, L. (1997) Curr. Biol. 7, R440–R442. 19. Cavicchi, S., Guerra, V., Natali, V., Pezzoli, C. & Giorgi, G. (1989) J. Evol. Biol. 36. Pletcher, S. D. & Curtsinger, J. W. (1998) Evolution 52, 454–464. 2, 235–251. 37. Carey, J. R., Liedo, P., Orozco, D. & Vaupel, J. W. (1992) Science 258, 457–461. 20. McColl, G., Hoffmann, A. A. & McKechnie, S. W. (1996) Genetics 143, 1615–1627. 38. Stearns, S. C. & Kaiser, M. (1996) Evolution 50, 795–806. EVOLUTION

Stearns et al. PNAS ͉ March 28, 2000 ͉ vol. 97 ͉ no. 7 ͉ 3313